ExSitu

EU AI Act Enforcement Is No Longer Theoretical: What Has Happened Since August 2

On August 2, 2026, enforcement of the EU AI Act's transparency obligations actually began. Google has already paid a fine and X is under investigation. What it means for an Israeli company with users in Europe.

2 min read

three european flags flying in front of a building
Unsplash · Ezgi Deliklitas

On August 2, 2026, enforcement of the transparency obligations in the EU Artificial Intelligence Act (EU AI Act) effectively began — not a future announcement, but actual enforcement. The European Commission's AI Office, the competent national authorities and the European Data Protection Supervisor are already operating under the new law.

What Is Already Required in Practice

Chatbots must identify themselves as automated systems, deepfakes must be labeled, and content generated or edited by a machine must carry a machine-readable marking that enables automatic detection. Providers of general-purpose AI models (GPAI) — like the ones most enterprise tools are built on — must document and share information with authorities and downstream providers, publish a summary of their training data, and implement a copyright policy.

The penalties are not symbolic: up to EUR 15 million or 3% of global annual turnover — whichever is higher.

Who Is Already in the Arena

Enforcement has not stayed on paper. Google previously paid a EUR 1 billion fine under other European laws (July 2026), and the X platform has been under investigation since January 2026 under the Digital Services Act (DSA) over its Grok tool. These are not warnings — they are open cases against some of the largest technology companies in the world.

More than 180 organizations have voluntarily signed the code of practice on transparency of AI-generated content — a tool designed to give companies a pre-agreed path to demonstrating compliance, rather than waiting for a regulator to define it for them.

Why This Concerns an Israeli Company

The EU AI Act applies based on where the user is, not where the company is. An Israeli company with a chatbot on its website and users or customers in the European Union — including an internal support tool used only by employees in Europe — falls within its scope.

"The fact that X is being investigated under two European laws at once — the DSA and the AI Act — shows that the authorities are not waiting for a full legal precedent before opening a case. Anyone who assumes enforcement will be slow is mistaken."

What This Means for Your Organization

  • Check right away whether your website or systems include a chatbot that carries the company's name — including one deployed through an external vendor without going through IT, and even one used only internally for employee support. This is no longer a recommendation for the future: enforcement actually began on August 2, 2026, and the potential fine reaches EUR 15 million or 3% of global annual turnover, whichever is higher.
  • If you have users or customers in the European Union — even through an internal support tool used only by employees there — make sure the statement "you are talking to an artificial intelligence system" appears clearly at the start of the interaction, not buried in terms of use that nobody reads. This check takes half an hour, and the consequences of skipping it are financial and immediate.
  • Contact your model provider — OpenAI, Anthropic, Google and others — and find out what compliance documentation they provide for the obligations that apply to you as an enterprise user (deployer): machine-readable content labeling, a training data summary and a copyright policy. Most organizations cannot implement this on their own, which is exactly why it is worth asking the vendor explicitly now, before a regulator asks you.
  • The case of X proves that European authorities are willing to open parallel investigations into the same company under several laws — in this case both the DSA and the AI Act. Do not assume enforcement will be gradual or lenient toward companies too small to be noticed — if it has not reached them yet, that is not because it will not.
  • Consider joining the voluntary code of practice on transparency of AI-generated content, or at least reviewing its requirements in depth. It is a pre-agreed path to demonstrating compliance, with more than 180 organizations already signed on — easier to rely on as an internal baseline than waiting for a future regulatory interpretation that will arrive only after the damage is done.

Sources